International Journal of Applied Engineering Research

ISSN 0973-4562 Volume 9, Number 21 (2014) pp. 11151-11162

© Research India Publications
http://www.ripublication.com

EMG Signal Features Extraction of Different Arm

Movement for Rehabilitation Device

M. H. Jali1,‘I.M. Ibrahim2, M. F. Sulaima3, T. A. Izzuddin4, W. M. Bukhari5

Faculty of Electrical Engineering, UniversitiTeknikalMalaysia Melaka, Malaysia

1
mohd.hafiz@utem.edu.my, 2iffamastura@gmail.com, 3fani@utem.edu.my,
4
tarmizi@utem.edu.my, 5bukhari@utem.edu.my

Abstract

Rehabilitation device is used as an exoskeleton for people who had failure of

their limb. Arm rehabilitation device may help the rehab program to who
suffer from arm disability. The device used to facilitate the tasks of the
program should improve the electrical activity in the motor unit and minimize
the mental effort of the user. Electromyography (EMG) is the techniques to
analyze the presence of electrical activity in musculoskeletal systems. The
electrical activity in muscles of disable person is failed to contract the muscle
for movements. To prevent the muscles from paralysis becomes spasticity the
force of movements should minimize the mental efforts. To minimize the used
of mental forced for disable patients, the rehabilitation device should analyze
the surface EMG signal of normal people that can be implemented to the
device. The signal is collected according to procedure of surface
electromyography for non-invasive assessment of muscles (SENIAM). The
EMG signal is implemented to set the movements’ pattern of the arm
rehabilitation device. The filtered EMG signal were extracted for features of
Standard Deviation(STD), Mean Absolute Value(MAV), Root Mean
Square(RMS) in time-domain. The extraction of EMG data is important to
have the reduced vector in the signal features with less of error. In order to
determine the best features for any movements, several trials of extraction
methods are used by determining the features that can be used in classifier.
The accurate features can be appliedin future works of rehabilitation control
system in real-time and classification of the EMG signal.

Keyword – Electromyography, Features Extraction, Time-domain Analysis,

Arm Rehabilitation Device, Upper Limb, Bilateral Movement

26864 - IJAER
11152 M. H. Jali et al

I. INTRODUCTION
Human support system is known as an endoskeleton. Endoskeleton plays a role as a
framework of the body which is bone. Our daily movements are fully depends on the
functionality of our complex systems in the body. The disability one or more of the
systems in our body will reduce our physical movements. Exoskeleton device is
known as rehabilitation device to facilitate disable person. The functionality of the
rehabilitation device has to smooth as the physical movement of normal human.
People who have temporary physical disability have the chances to recover. The
rehabilitation programs provide the suitable plan for conducting the nerve and
stimulate the muscles. Nowadays, rehabilitation program are using rehabilitation
devices in their tasks. The functionality of devices depends on muscle contraction.
Electromyogram studies help to facilitate the effectiveness of the rehabilitation device.
The technique of measuring electrical activity that produced by muscles during
rest or contractions known as electromyography (EMG). The electrical signal
generates from the brain and sends to the muscles via motor neuron. The EMG could
detect the dysfunctional of the muscles or failure in signals transmission from nerves
to muscles. The failure of sending the electrical signal from the brain to the
conducting nerves requires electrical stimulation from the external source to muscles.
Electrodes are used for signal detection of electrical activity in muscles. The study of
this electrical activity is important for combination of electromyogram into the
rehabilitation device.
The rehabilitation device is a tool that used to help the movements for daily life
activities of the patients who suffer from the failure of muscle contractions, due to the
failure of the muscles contractions the movements is limited. The ability of the
patients to do the tasks in the rehabilitation programs need to be measured. The
rehabilitation programs have to assure whether the tasks will cause effective or bring
harm to the patients[2].
Historically, the rehabilitation tasks have been avoided due to a belief that it
would increase spasticity [1]. In this research, the analysis of the data will be focusing
on upper limb muscles contraction consisting of biceps muscles only. The experiment
is limited to the certain of upper limb movements that use in training. EMG is a
division of bio signal; the bio signal analysis is the most complex analysis. Thus, the
signal analysis is a complicated process that has to be through many phases of
analysis [3].Therefore, the challenge of this study to assure that the signal processing
is conducted properly to overcome the environment noise during data collection [11].

II. SIGNAL PROCESSING

The output of the system commonly is defined by the input. The output is the main
achievement of the system. However, the EMG input is always a raw signal that needs
to be process into several plants until it can produce the desired output. EMG signal is
one type of bio signal that contains lot of noise from many factors. The noise may
come from the skin at the electrode placement, type of electrode that may use for data
collecting and also the environment. There are various type of size and pattern of
electrodes, depends on the muscle area would like to detect and the thickness of the
EMG Signal Features Extraction of Different Arm Movement 11153

skin layer. The EMG noise signal can be reduced by choosing the gelled electrode
which were Silver –Silver Chloride(Ag-AgCl) as the substance. Ag-AgCl introduces
less electrical noise into the measurements [13]. However, these noise should be
eliminate to have a better signal for analysing purpose, the noise can be eliminate by
many factors such as control environment, electrode placements and signal
acquisition circuitry and configuration.[1][2] As for the features extraction there were
various types in terms of frequency domain, time frequency domain and time scale
domain. These types are chose based on the purpose of the analysis of the signal. In
signal conditioning for time domain analysis, there is some features that can be
applied. These features are described briefly as follows:

i) Root-Mean-Square (RMS)
RMS is the square root of the mean over time of the square of the vertical distance of
the graph from the rest state. It much related to the constant force and non-fatiguing
contraction of the muscle. In most cases, it is similar to standard deviation method.
1
= ()

ii) Mean Absolute Value(MAV)

Mean absolute value is calculated by taking the average of the absolute value of EMG
signal. Since it represents the simple way to detect muscle contraction levels, it
becomes a popular feature for myoelectric controlled applications.
1
= |( − ̅ )|

iii) Standard Deviation(STD)

Standard deviation is a measure of the dispersion of a set of data sample from its
mean. The more spread apart the data, the higher the deviation. Standard deviation is
calculated as the square root of variance.

1
= ( − ̅)
−1

III. DATA ACQUISITION IN SEMG

Surface EMG is the technique that had been applied in this study for recording the
electrical activity in the muscles by the electrodes. The sEMG provide an easy, safe
and non-invasive method that allow the quantification of the muscles’ energy.
Recording the signal are depends on several criteria that need to be considered before
the signal is recorded. According to Zahak Jamal et al.on his studies, the electrode
placements are an important issue that need to be considered, the placements is
depends on the type of the muscles as the muscles at each limb is different. The
quality of the sEMG signal that generates from the muscles’ energy can be maximizes
11154 M. H. Jali et al

by placing the electrodes at the belly of the muscles not at or near the tendon and
motor unit. If the electrodes is bipolar configuration(Fig 1.), which have three
detecting surfaces, both electrode need to be placed nearly to each other within 1-2cm
[4]. These two detecting surfaces are connected to differential amplifier while another
one electrode is placed at the reference(bone).

Fig 1. Bipolar configuration

The aim for this study to have the EMG signal for analysis purpose by reduce all
the unwanted noise from the electrode, hardware and the environment to maximize
the quality of the EMG signal. There have a lot of noise factor that can be categorized
as follows;
i) Causative Factors: This is the direct affected on signals.
a. Extrinsic – The signal get affected by the electrode placement, structure, surface
detection, distance between electrode(bipolar configuration), and location of the
surface detection on the skin.
b. Intrinsic – It may due to physiological, biochemical and anatomical factors.
ii) Intermediate Factors: This is the phenomena of influencing by one or more
causative factors.
iii) Deterministic Factors: This is influenced by the intermediate factors.

By reducing all these factor of noises, the quality of EMG signal will be enlarge
and the analysis of the information will be less of error and easy to obtain. However,
the precautions of handling the hardware and the electrode are needed. Acquisition
data play the important role to maximizing the quality of the EMG signal, such as
minimizing the distortion in EMG signal, using any filtering tools are not
recommended. In terms of signal-to-ratio(SNR) the information that carried in SNR
should contain the maximum information of EMG signal[3][10]. Moreover, the
quality of EMG signal is affected by the environment,has to set by minimizing the
noise factor, this method also known as control environment. The ambient noises
sourced by electromagnetic device such as radio transmission, fluorescent lights and
power line interference from electrical wires[10][4]. These ambient noise and motion
artifact can be reduced by the proper electrode placements, circuit configuration and
control environment. The signal recorded in Fig 2. shows the appearance of electrical
activity in muscles cannot be differentiate by any muscles contraction, because of not
EMG Signal Features Extraction of Different Arm Movement 11155

following the criteria to reduce the noises and the electrode placements detection
surface range.

Arm movements from 90deg to 150deg

0

-20

-40
amplitude response

-60

-80

-100

-120

-140
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
t(sec)

Fig.2.The effects of placement the bipolar electrodes not within 1-2cm and
without environment control (ambient noise).

IV. EXPERIMENTAL SETUP

The experiment’s environment is in a room with low lighting especially the
fluorescent light, any electromagnetic devices is away from the experiment equipment
and the environment is in soundless room. Then, the experimental is set up with the
subject sit on the chair while the hand is on the table. The subject has to complete the
task of lift up the dumbbell with 2.27kgs of weigh in Figure 5 for 10 times. Mostly,
the EMG signal is obtained after several trials of the movements. Normally, the
appearance of EMG signal is chaos and noisy depends on the type of electrodes also
the noise factor. To simplify the difference of amplitude response’s motion, the
dumbbell is functioned to amplify the amplitude in analyzing the electrical activity
during rest and contract. The rehabilitation devices(white in colour on Fig. 3) helps to
keep the position of the elbow joint and the wrist joint in line without being affected
by the weight of the dumbbell during lift up motion. These movements are specified
from angle of 0°(arm in rest position), up to 120°(arm is fully flexion) [12][14]. The
experiment will go through several phases as explained in the next section.
11156 M. H. Jali et al

Fig. 3.Subject Is Set-Up with Arm Rehabilitation Assistive Device for

Experiment

Fig.4.Simulation of Subject Has To Lift Up the Dumbbell 2.27 Kgs of Weight

1) Phase 1: Skin Preparation and Electrode Placements.

The placement of the electrode is followed after skin preparation. The preparation of
the skin is needed before placing the electrodes. The subject’s skin has to shave using
small electrical shaver and cleaned with sterile alcohol swabs saturated with 70%
Isopropyl Alcohol as shown in Fig 5(a). This step is to be taken for minimising the
noise and to have a good contact with the electrodes of the skin. The skin has to be
clean from any contamination of body oil, body salt, hair and the dead cells. The
preparation of skin can be done by wipe the alcohol swab into the area of skin that
electrode placement to be applied. The placements of the electrode have to be at the
belly of the muscles not in the tendon or motor unit. Electrode leads and gelled
electrode (Fig 5(b)) is used to collect the data. The combination of Arduino Mega and
the OLIMEX shield, simplified the hardware. As the EMG signals are too sensitive
EMG Signal Features Extraction of Different Arm Movement 11157

towards magnetic devices, the neatness of the hardware is needed. The neatness
ensured the electrode leads and the detecting surface intersects most of the same
muscle on subject as in Fig 6(b) at biceps, and as a result, an improved superimposed
signal is observed. Reference electrode has to be at the body where no muscles exist
as the ground signal, for this experiment it placed at wrist on the unaffected
hand(refer Fig 6(b)).

(a) (b)

Fig. 5.The Alcohol Swab(a) and the Gelled Electrode(b) That Used in This
Experiment.

2) Phase 2: sEMG Signal Acquisition

Biceps

Biceps
(a) (b)

Fig.6. The Biceps Brachii Muscles For Electrode Positions(a), The Electrode
Placements On Subject Skin(b)
11158 M. H. Jali et al

Gelled electrode has two leads detecting surfaces and one lead as reference
electrode. These two detecting surface need to be placed in range of less than 1-2 cm
from each of it at the biceps muscle as in Fig 6(b). The gelled electrode for the blue
lead and red lead are placed at the biceps skin for detection purpose, meanwhile the
gelled electrode with black lead is placed at area with no muscles exist and unaffected
hand.[1] The grounded electrode hand is placed at the leg to have a better grounding
purpose. From Figure 8 these EMG signals will go through into high voltage
protection to protect any electrical surge that may bring harm to the user and the
hardware. These signals are available for signal conditioning process as the next step
to be taken. The circuit that be used in this experiment is provided by the OLIMEX
shield. The signals via the OLIMEX shield through the analogue circuit are very weak
the voltage is around 10μV, full of noise and contain primarily 60Hz of frequency.
The signals are next need to be filter and rectify, reducing the noise and amplifying
the signals for analysis purpose. After the signals is filtered and amplified the signals
is sent to the digital board and yet ready to display into the MATLAB software.

Fig.7.The Electrical Diagram of Electrode Sensor, High Voltage Protection and

Frequency Rejection.

3) Phase 3: Signal Conditioning

The signals from the board are gathered in the SIMULINK software. Signals from the
OLIMEX shield need to be filtered by using the peak notch filter and the high pass
filter. Peak notch filter is used for filtering the interference from the line AC and radio
frequency from the devices frequency nearby. Meanwhile, the high pass filter is used
for placed the baseline of EMG signals display in the scope(SIMULINK) at the zero
line, also known as DC offset. Block diagram for collecting the EMG signal is
showed in Figure 9. The raw EMG signals is next to filter by using Butterworth filter.
The Butterworth filter is set to 5th order and 10/1000 Hz frequency. The filtered signal
is accessed for features extraction. This EMG signal evaluated in time domain
features. The analysis of this study soon to be applies in real-time application. Thus,
the time-domain suits this study for better future work. These features are Root-Mean-
EMG Signal Features Extraction of Different Arm Movement 11159

Square(RMS), Mean Absolute Value(MAV), Standard Deviation(STD),

Variances(VAR), and Zero Crossing(ZCS). This statistical information of the
features evaluations were run in the MATLAB R2013a. The signals that gathered
from the SIMULINK is next to be apply for features extraction. The features are
extracted using MATLAB 2013a programming.

V. RESULTS AND DISCUSSIONS

The filtered signal goes into features extraction process. The features extraction is set
into five types which were RMS, MAV, STD, ZCS and VAR. These features are
extracted from each movement of individual EMG signal. The data of the features is
accumulated in Table 1. The suitable features that may use in classifying the
movement are MAV, RMS and STD. ZCS and VAR value are not applicable in
classifying the features. ZCS produced the whole numbers, meanwhile the VAR
produced the scientific number that smaller in range of 10-6. However, the ZCS and
VAR features may be used as it is, in its own class of features. The results are various
except the feature is preset.

TABLE I The Featuresof Degree Movements by Angle.

Subjects Angle MAV RMS STD ZCS VAR

S1 60˚ 0.001903 0.001981 0.001329 50 1.68E-06
1.61E-03 0.001777 0.001329 48 1.99E-06
1.42E-03 1.75E-03 0.001325 44 0.001325
90˚ 0.00232 0.0027 0.0022 58 2.66E-06
0.00231 0.00267 0.00219 43 2.15E-06
0.00228 0.00254 0.00215 37 2.189E-06
120˚ 0.00318 0.00363 0.00311 39 3.13E-06
0.00317 0.00365 0.0031 41 2.71E-06
0.00317 0.00339 3.16E-03 51 3.71E-06
S2 60˚ 0.00174 0.00219 1.42E-03 65 9.50E-07
0.00165 0.00189 1.38E-03 65 9.75E-06
0.00164 0.00188 1.21E-03 31 1.76E-06
90˚ 0.00244 0.00286 0.00238 81 2.35E-06
0.00237 0.00285 0.00231 85 2.62E-06
0.00232 0.00277 0.00228 54 1.62E-06
120˚ 0.00315 3.99E-03 0.00316 13 3.56E-06
0.00316 3.98E-03 0.00318 33 3.93E-06
0.00315 0.00384 0.00314 45 3.42E-06
S3 60˚ 9.75E-04 0.0019 0.00124 27 8.34E-07
9.15E-04 0.00162 0.00133 19 1.22E-06
9.13E-04 1.61E-03 1.10E-03 19 8.38E-07
90˚ 0.002 0.00239 0.00209 23 1.24E-06
0.00189 0.00231 0.00207 17 1.93E-06
11160 M. H. Jali et al

0.00186 0.00228 0.00205 41 1.75E-06

120˚ 0.00282 0.00318 0.00314 31 2.49E-06
0.00291 0.00316 0.00307 50 2.53E-06
2.98E-03 0.00315 0.00302 23 2.60E-06
S4 60˚ 0.00161 0.00198 0.00133 15 1.89E-06
0.00154 0.00185 0.00129 19 1.46E-06
0.00141 1.73E-03 0.00127 20 2.00E-06
90˚ 0.00233 0.00246 2.44E-03 30 2.21E-06
0.00231 0.00243 0.00237 25 2.75E-06
2.23E-03 0.00242 0.0026 27 2.06E-06
120˚ 0.0033 0.00391 0.00317 34 3.05E-06
0.0032 0.00382 0.00313 36 3.16E-06
0.00327 0.00378 3.11E-03 37 3.01E-06

ZCS and VAR formed the whole and 10-6 scientific numbers that would not
achievable to be class into the classifier because of the distance value of features
itself. Data from Table 1 is plotted into scatter graph in Fig. 8. Sample 1 up to sample
4 be owned by movement of 60˚, sample 5 to sample 8 is belong to movement of 90˚
and sample 9 to sample 12 is fit in movement 120˚. The green, yellow and red outline
colour in Fig. 8shows that the scatter features is obviously been clustered by its angle
of movements’ class. Even though the clustered movements is scattered in its class,
the objective of this study is achievable. The features of movements are belongs in
their movement group. The learning of the features is continued in data classification
[15][16].The time-domain features for STD, MAV and STD shows in Fig. 8 useful for
future works such as classification of the features. The classification of time-domain
features provide reliable information that might be use with combination these three
features for research in many type of classification technique.

0.004
AMPLITUDE (V)

0.003
STD

0.002 MAV

RMS
0.001

0
0 5 10 15
SAMPLES NUMBER

Fig. 8.The Scatter Graph of Features versus Amplitude Voltage.

EMG Signal Features Extraction of Different Arm Movement 11161

VI. CONCLUSION
In this study has further discussed on the noise and the environment that affects the
EMG signal during data collection. The surface electrodes contain lot of noise
compared to needle insertion procedure, noise from the skin itself and the
environment have been discussed in this study for better understanding to get a better
quality of EMG signals.
The objectives of this study is to extract the features have successfully obtained
from the experiment are MAV, STD, RMS, ZCS and VAR. On the other hand, not all
the features can be classified into the class of several degrees of movements. The
suitable features for classification are MAV, STD and RMS. The features are obtained
by signal processing phase. The features of time-domain are extracted by using
programming software use in MATLAB software. MAV, STD and RMS verified in
scatter graph that it can be use for several movements of classification in future works.
The uses of these features might be separately or combination among the all three.

ACKNOWLEDGMENT
The authors would like to thanks UniversitiTeknikal Malaysia Melaka (UTeM) and
Ministry of Education, Malaysia for the financial supports given through Research
Grant.

REFERENCES

[1] B. B, "Adult Hemiplegia: Evaluation and treatment.," Oxford, Butterworth-

Heinemann., 1990.
[2] S. D. C. G. C. Louise Ada, "Strengthening interventions increase strength and
improve activity after stroke: a systematic review," Australian Journal of
Physiotherapy, vol. 52, pp. 241-248, 2006.
[3] J. Muthuswamy, "Biomedical Signal Analysis," In Standard Handbook Of
Biomedical Engineering And Design, Mc-Graw Hill, 2004, pp. 18.1 - 18.30.
[4] M. Z. Jamal, "Signal Acquisition Using Surface EMG and Circuit Design
Considerations for Robotic Prosthesis," in Computational Intelligence in
Electromyography Analysis –A Perspective on Current Applications and Future
Challenges, InTech, 2012, pp. 427-448.
[5] S.-P. L. Sang-Hui Park, "EMG Pattern Recognition Based on Artificial
Intelligence Techniques," IEEE Transactions On Rehabilitation Engineering, vol.
6, no. 4, 1998.
[6] C. L. a. P. P. Angkoon Phinyomark, "A Novel Feature Extraction for Robust
EMG Pattern Recognition," Journal Of Computing, vol. 1, no. 1, pp. 71-80, 2009.
[7] K. Mahaphonchaikul, "EMG Signal Feature Extraction Based on Wavelet
Transform".
[8] P. M. S. K. Mr. Santosh S. Patil, "Feature Extraction of MES (MyoElectricSignal)
using Temporal approach for prosthetic arm control," International Journal of
Electronics and Computer Science Engineering.
11162 M. H. Jali et al

[9] L. Oujamaa, "Rehabilitation of arm function after stroke. Literature review,"

Annals of Physical and Rehabilitation Medicine, vol. 52, pp. 269-293, 2009.
[10] M. B. I. Reaz, "Techniques of EMG signal analysis: detection, processing,
classification and applications," Cyberjaya, Selangor,
www.biologicalprocedures.com, 2006, pp. 11-35.
[11] M. H. Jali, M. F. Sulaima, T. A. Izzuddin, W. M. Bukhari, M. F.
Baharom.,“Comparative Study of EMG based Joint Torque Estimation ANN
Models for Arm Rehabilitation Device”, International Journal of Applied
Engineering Research (IJAER), vol. 9(10), pp. 1289-1301, 2014.
[12] MH Jali, TA Izzuddin, ZH Bohari, H Sarkawi, MF Sulaima, MF Baharom, WM
Bukhari, "Joint Torque Estimation Model Of sEmg Signal For Arm Rehabilitation
Device Using Artificial Neural Network Techniques", ARPN Journal of
Engineering and Applied Sciences, vol 9(7), pp 1170 - 1176, 2014
[13] M.H.Jali, ‘I.M. Ibrahim, M.F.Sulaima, W.M Bukhari, T.A.Izzuddin and M.N.M.
Nasir, “Features Extraction of EMG Signal using Time Domain Analysis for Arm
Rehabilitation Device”, International Conference on Mathematics on Mathematics,
Engineering and Industrial Applications (ICoMEIA), 2014.
[14] M.H.Jali, T.A.Izzuddin, Z.H.Bohari, M.F.Sulaima, H.Sarkawi, “Predicting EMG
Based Elbow Joint Torque Model Using Multiple Input ANN Neurons for Arm
Rehabilitation “,UKSim-AMSS 16th International Conference on Modelling and
Simulation, 2014
[15] M.H.Jali, Z.H.Bohari, M.F.Sulaima, M.N.M.Nasir and H.I.Jaafar, “Classification
of EMG Signal Based on Human Percentile using SOM”, Research Journal of
AppliedSciences, Engineering and Technology (RJASET), 2014.
[16] M.H.Jali, I.M.Ibrahim, Z.H.Bohari, M.F.Sulaima, M.N.M.Nasir, ”Classification
Of Arm Movement Based On Upper Limb Muscle Signal For Rehabilitation
Device”, Journal of Theoretical and Applied Information Technology (JATIT),
2014.